Research map / Architectures and efficiency / Attention and transformers
Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs
Ngoc Bui, Shubham Sharma, Simran Lamba, Saumitra Mishra, Rex Ying
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add forward KL distillation loss (ablation) | TRIM-KV · AIME24 (budget 4096) | pass@1: 72.1 -> 75.8 | Improved |
| add next-token prediction loss (ablation) | TRIM-KV · AIME24 (budget 4096) | pass@1: 72.5 -> 75.8 | Improved |
| add memory capacity loss (ablation) | TRIM-KV · AIME24 (budget 4096) | pass@1: 42.9 -> 75.8 | Improved |
| evict with TRIM-KV instead of SnapKV at 32768 budget | Qwen3-4B-Instruct · LongMemEval_S | accuracy (%): 27.6 -> 44.8 | Improved |
| evict with TRIM-KV instead of SnapKV at 8192 budget | Qwen3-4B-Instruct · LongMemEval_S | accuracy (%): 15.8 -> 34.0 | Improved |
| evict with TRIM-KV instead of LocRet | Phi3-mini-128K · LongBench-V2 | average accuracy (%): 28.03 -> 34.09 | Improved |
| evict with TRIM-KV instead of LocRet | Phi3-mini-128K · LongBench | average relative change (%): -4.82 -> -2.54 | Improved |
| use TRIM-KV instead of R-KV at 2048 budget | Qwen3-4B · LongProc Pseudo-to-Code (2k) | accuracy (%): 2.5 -> 18.0 | Improved |
| use TRIM-KV eviction instead of full cache | Qwen3-4B · LongProc Countdown (2k) | accuracy (%): 90.0 -> 93.5 | Improved |
| use TRIM-KV eviction instead of full cache | Qwen3-4B · AIME24 | pass@1: 65.5 -> 75.8 | Improved |
| use TRIM-KV eviction instead of full cache | Qwen3-4B · LongProc Pseudo-to-Code (2k) | accuracy (%): 25.0 -> 18.0 | Hurt |
| apply TRIM-KV eviction on incompressible retrieval | Qwen3-4B-Instruct · SCBench Retr.KV | score: 66.00 -> 0.00 | Hurt |
Closest papers by meaning
- SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget (ICLR 2025 · similarity 0.93)
- ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference (NeurIPS 2025 · similarity 0.91)
- RazorAttention: Efficient KV Cache Compression Through Retrieval Heads (ICLR 2025 · similarity 0.91)
- Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference (ICML 2024 · similarity 0.91)
- SCBench: A KV Cache-Centric Analysis of Long-Context Methods (ICLR 2025 · similarity 0.91)
- Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference (NeurIPS 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
